Related Experiment Video
Updated: Jul 3, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.4K
An interpretable machine learning model based on optimal feature selection for identifying CT abnormalities in
Yuling Pan1,2, Mengqi Wei3, Mengyuan Jin4
1School of Laboratory Medicine, Hubei University of Chinese Medicine, 16 Huangjia Lake West Road, Wuhan, 430065, China.
Eclinicalmedicine
|April 17, 2025
Summary
This study developed an interpretable machine learning model using routine lab data to predict brain lesions in mild traumatic brain injury (mTBI) patients, reducing unnecessary CT scans and radiation exposure.
Area of Science:
- Medical Informatics
- Machine Learning in Healthcare
- Clinical Decision Support Systems
Background:
- Minor head trauma is a common reason for emergency department visits.
- Accurate prediction of mild traumatic brain injury (mTBI) with abnormal brain lesions is crucial for appropriate CT scan use.
- Minimizing radiation exposure and ensuring timely treatment are key goals in mTBI management.
Purpose of the Study:
- To develop and validate an interpretable machine learning (ML) model for predicting CT abnormalities in mTBI patients.
- To utilize routine laboratory data for guiding clinical decisions on CT scan utilization.
- To enhance the accuracy and efficiency of mTBI diagnosis.
Main Methods:
- A multicentre study involving retrospective and prospective cohorts with 1,196 mTBI patients.
- Collection of 51 routine clinical laboratory characteristics from electronic medical records within 24 hours of admission.
- Training and validation of seven ML algorithms, with feature selection and interpretation using SHapley Additive exPlanations (SHAP) and decision curve analysis (DCA).
Main Results:
- The Gradient Boosting Classifier (GBC) model achieved high predictive performance (AUC 0.932 in derivation, 0.926 internal, 0.904 external validation).
- The optimized GBC model, using 21 key laboratory features, outperformed traditional biomarkers GFAP and PGP9.5 in prospective validation (AUC 0.885).
- SHAP analysis identified D-dimer, lymphocyte count, neutrophil count, and hematocrit as significant predictors of CT abnormalities.
Conclusions:
- An interpretable ML model using routine laboratory data can accurately predict CT abnormalities in mTBI patients.
- The developed model offers significant clinical net benefits, supporting informed decisions on CT scan use.
- This approach enhances diagnostic accuracy, reduces unnecessary radiation exposure, and improves patient care pathways for mTBI.
